1.
Congressional Budget Office
American Society of Health Economists Conference
Washington, D.C.
June 24, 2019
Geena Kim and Sean Lyons
Health, Retirement, and Long-Term Analysis Division
HISIM2—CBO’s New Health Insurance
Simulation Model

2.
1
CBO
Current Model Specification

3.
2
CBO
A health insurance unit (HIU) is the decisionmaking unit in HISIM2.
 A single person is his or her own HIU.
 Otherwise, an HIU is the set of individuals who could be covered by a family
plan (any plan that covers two or more people) if an employer were to offer that
plan.
An HIU collectively chooses the type of health insurance coverage in which to
enroll each of its members. People within the same HIU may not be eligible for
the same type of coverage and do not necessarily choose the same coverage
option.
HIUs make decisions by maximizing utility in a random utility model.
CBO models insurance characteristics by applying the plan’s deductible,
coinsurance, and maximum out-of-pocket limit to an HIU’s expected total health
care spending distribution.
 This allows us to more easily model policies that change the generosity of
insurance plans.
Overview of HIUs’ Behavior

4.
3
CBO
The utility of each alternative depends on the HIU’s total income, health care
spending (including premiums, an out-of-pocket spending distribution, and any
applicable subsidies, taxes, and mandate penalties), financial risk, and
unobserved factors.
 Out-of-pocket spending is determined by the health status of each member of
the HIU and by the cost-sharing characteristics of the insurance plan
(deductible, coinsurance, and maximum out-of-pocket limit).
 Utility is specified to decrease as financial risk increases.
Many utility function parameters are estimated by minimizing the difference
between the coverage predictions from the model and coverage targets for the
base year of data.
 Coverage targets are CBO’s preliminary estimates from individual data sources
of the actual number of people with a particular coverage status.
The remaining parameters are set on the basis of CBO’s assessment of the
research literature.
HIUs’ Utility

5.
4
CBO
Employment-based coverage is coverage offered by a current or former employer—either one’s own or a family member’s. Firms are restricted to offer only one type of plan: a high
deductible health plan (HDHP), health maintenance organization (HMO), or preferred provider organization (PPO). Nongroup coverage is coverage that a person purchases directly
from an insurer or through a health insurance marketplace, rather than through an employer. Plans in the nongroup market are categorized into tiers (which are named after metals) on
the basis of their actuarial value (the percentage of total average costs for covered benefits for which a plan pays). “Bronze” plans are those with an actuarial value of 60 percent,
“silver” plans are those with an actuarial value of 70 percent, and “gold” plans are those with an actuarial value of 80 percent.
HIUs select the type of insurance for each person in the unit:
 Employment-based coverage—single or family
 Nongroup coverage in the marketplaces—bronze, silver, or gold
 Nongroup coverage outside the marketplaces—bronze, silver, or gold
 Medicaid
 Children’s Health Insurance Program (CHIP)
 Medicare
 None (Uninsured)
HIUs’ Insurance Options

6.
5
CBO
The utility for HIU 𝑖𝑖 from alternative 𝑛𝑛 with multiple members 𝑗𝑗 = 1, … , 𝐽𝐽𝑖𝑖 (𝑈𝑈𝑖𝑖 𝑖𝑖), is the predicted value
that the HIU places on the alternative, 𝛽𝛽𝑉𝑉𝑖𝑖 𝑖𝑖, and an idiosyncratic unobservable component, 𝜀𝜀𝑖𝑖 𝑖𝑖.
𝑈𝑈𝑖𝑖 𝑖𝑖 = 𝛽𝛽𝑉𝑉𝑖𝑖 𝑖𝑖 + 𝜀𝜀𝑖𝑖 𝑖𝑖
𝑉𝑉𝑖𝑖 𝑖𝑖 = 𝑦𝑦𝑖𝑖 − 𝐶𝐶𝑖𝑖 𝑖𝑖 − 𝐸𝐸s 𝐻𝐻𝑖𝑖 𝑖𝑖 −
1
2
𝜌𝜌𝑛𝑛 𝑉𝑉𝑉𝑉𝑟𝑟s 𝐻𝐻𝑖𝑖 𝑖𝑖 +Δ𝑖𝑖 𝑖𝑖 𝑦𝑦𝑖𝑖, 𝑎𝑎1, … , 𝑎𝑎𝐽𝐽𝑖𝑖, 𝑒𝑒1, … , 𝑒𝑒𝐽𝐽𝑖𝑖
𝑦𝑦𝑖𝑖 is income for HIU 𝑖𝑖, 𝑎𝑎𝑗𝑗 is the age of member 𝑗𝑗 ∈ 𝑖𝑖, and 𝑒𝑒𝑗𝑗 is eligibility for coverage for member 𝑗𝑗 ∈ 𝑖𝑖.
𝐶𝐶𝑖𝑖 𝑖𝑖 is the cost to HIU 𝑖𝑖 of alternative 𝑛𝑛, which includes premiums and applicable individual-mandate
penalties.
𝐸𝐸s 𝐻𝐻𝑖𝑖 𝑖𝑖 and 𝑉𝑉𝑉𝑉𝑟𝑟s 𝐻𝐻𝑖𝑖 𝑖𝑖 are the expectation and variance, respectively, of HIU 𝑖𝑖’s out-of-pocket health
care spending for each alternative 𝑛𝑛 where the expectation and variance are taken over health
states 𝑠𝑠. Out-of-pocket health care spending for the uninsured alternative is capped at a value that
increases with income to capture uncompensated care and bankruptcy.
𝜌𝜌𝑛𝑛 is the coefficient of absolute risk aversion.
Δ𝑖𝑖 𝑖𝑖 is the alternative-specific constant for alternative 𝑛𝑛 and is allowed to vary with age, eligibility, and
income. When HIU size is one, Δ𝑖𝑖 𝑖𝑖 = 𝛿𝛿1𝑛𝑛 𝑦𝑦𝑖𝑖, 𝑎𝑎𝑗𝑗 , and when HIU size is greater than one,
Δ𝑖𝑖 𝑖𝑖 = ∑𝑗𝑗=1
𝐽𝐽𝑖𝑖
𝛿𝛿2𝑛𝑛(𝑦𝑦𝑖𝑖, 𝑎𝑎𝑗𝑗, 𝑒𝑒𝑗𝑗).
Utility Specification

7.
6
CBO
The unobservable component is modeled so that it implies a generalized nested logit (GNL) model. For notation
purposes, the GNL choice probability is:
𝑃𝑃𝑖𝑖 𝑖𝑖 =
∑ℓ 𝛼𝛼𝑛𝑛ℓ exp 𝛽𝛽𝑉𝑉𝑖𝑖 𝑖𝑖
1
𝜇𝜇ℓ ∑𝑘𝑘∈𝐵𝐵ℓ
𝛼𝛼𝑘𝑘ℓ exp 𝛽𝛽𝑉𝑉𝑖𝑖 𝑖𝑖
1
𝜇𝜇ℓ
𝜇𝜇ℓ−1
∑ℓ ∑𝑘𝑘∈𝐵𝐵ℓ
𝛼𝛼𝑘𝑘ℓ exp 𝛽𝛽𝑉𝑉𝑖𝑖 𝑖𝑖
1
𝜇𝜇ℓ
𝜇𝜇ℓ
In this equation, nests (employment-based, public, nongroup marketplace, nongroup off marketplace, and
uninsured) are indexed by ℓ and the set of choices in nest ℓ is denoted 𝐵𝐵ℓ.
CBO estimates most utility function parameters by fitting the model’s equations to the data but sets the value of
others on the basis of information from research and from experts on health insurance markets and this type of
model.
The alternative-specific constants (Δ𝑖𝑖 𝑛𝑛) and utility coefficient (𝛽𝛽) are estimated.
All remaining parameters are set:
 The coefficient of absolute risk aversion (𝜌𝜌𝑛𝑛) and the dissimilarity parameters (𝝁𝝁), which determine correlation
between choices within a nest, are set on the basis of the available evidence.
 The allocation parameters (𝜶𝜶) are set by calculating the fraction of individuals who would be covered under each
source of coverage from all the HIUs that are able to make a given choice.
Utility Specification (Continued)

8.
7
CBO
CBO estimates the alternative specific constants (Δ𝑖𝑖 𝑛𝑛) by finding values that minimize the difference between coverage
predictions from the model and coverage targets by type of insurance, age, and income (minimum distance estimation).
The utility coefficient, 𝛽𝛽, determines the degree to which HIUs’ decisions are predicted by the value function specification
and are estimated on the basis of how well the model fits the underlying data.
Specifically, 𝛽𝛽 is estimated via a maximum likelihood estimation in which the choice is insured versus uninsured.
The minimum distance estimation (MD) portion of the objective function is:
𝑀𝑀𝑀𝑀 𝜹𝜹𝒊𝒊, 𝛽𝛽 =
1
N
Γ 𝜹𝜹𝒊𝒊, 𝛽𝛽 ′Γ 𝜹𝜹𝒊𝒊, 𝛽𝛽
 Each element of Γ(⋅) is the model’s predicted aggregate coverage total differenced from our target.
The maximum likelihood estimation (LL) portion of the objective function is:
𝐿𝐿𝐿𝐿 𝜹𝜹𝒊𝒊, 𝛽𝛽 =
1
𝑁𝑁
�
𝑖𝑖
�
𝑗𝑗∈𝑖𝑖
𝑤𝑤𝑤𝑤𝑡𝑡𝑗𝑗 ⋅ 𝑃𝑃𝑗𝑗 𝑖𝑖 𝑖𝑖𝑖𝑖 𝜹𝜹𝒊𝒊, 𝛽𝛽 − �
𝑗𝑗∈𝑖𝑖
𝑤𝑤𝑤𝑤𝑡𝑡𝑗𝑗 ⋅ 𝑑𝑑𝑗𝑗 𝑖𝑖 𝑖𝑖𝑖𝑖
2
 𝑤𝑤𝑤𝑤𝑡𝑡𝑗𝑗 is person 𝑗𝑗′
s sample weight, 𝑑𝑑𝑗𝑗 𝑖𝑖 𝑖𝑖𝑖𝑖 is an indicator for whether or not the person is insured, and 𝑃𝑃𝑗𝑗 𝑖𝑖 𝑖𝑖𝑖𝑖(𝜹𝜹𝒊𝒊, 𝛽𝛽) is
the probability that the person is insured.
So, the objective function is:
min
𝜹𝜹𝒊𝒊,𝛽𝛽
𝑀𝑀𝑀𝑀 𝜹𝜹𝒊𝒊, 𝛽𝛽 + 𝐿𝐿𝐿𝐿(𝜹𝜹𝒊𝒊, 𝛽𝛽)
HIU Utility Parameters

9.
8
CBO
Calibration Estimates

10.
9
CBO
FPL = federal poverty level.
Calibration Results for Nongroup Coverage, 2015
Percent HIU = 1 HIU > 1
Description
Difference From
Target
Take-Up
Rate
Difference From
Target
Take-Up
Rate
Marketplace Bronze Subsidized FPL 0-69
Marketplace Bronze Subsidized FPL 70-250 <.01 7.2 <.01 7.6
Marketplace Bronze Subsidized FPL 251-400 <.01 7.0 <.01 10.2
Marketplace Bronze Subsidized FPL 401+
Marketplace Silver Subsidized FPL 0-69
Marketplace Silver Subsidized FPL 70-138 age < 30 <.01 18.3 -0.01 35.2
Marketplace Silver Subsidized FPL 70-138 age 30-50 <.01 53.8 <.01 30.1
Marketplace Silver Subsidized FPL 70-138 age > 50 <.01 64.8 -0.01 57.2
Marketplace Silver Subsidized FPL 139-250 age < 30 <.01 14.4 <.01 55.8
Marketplace Silver Subsidized FPL 139-250 age 30-50 <.01 31.6 <.01 38.8
Marketplace Silver Subsidized FPL 139-250 age > 50 <.01 57.9 <.01 60.4
Marketplace Silver Subsidized FPL 250-400 <.01 7.6 <.01 12.5
Marketplace Silver Subsidized FPL 401+
Marketplace Gold Subsidized FPL 0-69
Marketplace Gold Subsidized FPL 70-400 <.01 2.1 <.01 4.2
Marketplace Gold Subsidized FPL 401+
Marketplace Bronze Unsubsidized <.01 1.3 0.01 0.2
Marketplace Silver Unsubsidized <.01 1.7 <.01 0.3
Marketplace Gold Unsubsidized <.01 0.7 0.01 0.1
Nongroup Outside-the-Marketplaces Bronze <.01 3.4 <.01 0.6
Nongroup Outside-the-Marketplaces Silver <.01 12.0 <.01 2.5
Nongroup Outside-the-Marketplaces Gold <.01 1.1 <.01 0.3

11.
10
CBO
Calibration Results for Employment-Based Coverage, 2015
Percent HIU = 1 HIU > 1
Description
Difference
From Target
Take-Up
Rate
Difference
From Target
Take-Up
Rate
PPO—Family, HIU > 1 <.01 53.8
HMO—Family, HIU > 1 <.01 48.8
HDHP—Family, HIU > 1 <.01 47.2
Nondependent Child, HIU > 1 <.01 22.6
PPO—Single, HIU = 1 <.01 77.7 <.01 24.6
HMO—Single, HIU = 1 <.01 85.2 <.01 24.4
HDHP—Single, HIU = 1 <.01 85.1 <.01 48.2
Employment-Based Coverage + Medicaid
Children <.01 22.4

12.
11
CBO
Calibration Results for Public Coverage, 2015
Percent HIU = 1 HIU > 1
Description
Difference
From Target
Take-Up
Rate
Difference
From Target
Take-Up
Rate
Medicaid Children <.01 90.2
Adults Made Eligible for Medicaid by the
Affordable Care Act <.01 58.5 <.01 70.6
Adults Otherwise Eligible for Medicaid <.01 90.0 <.01 90.6
CHIP <.01 48.5
Medicare With Offer of Employment-Based
Coverage <.01 59.5 <.01 54.5

13.
12
CBO
FPL = federal poverty level.
Calibration Results for the Uninsured, 2015
Percent HIU = 1 HIU > 1
Description
Difference
From Target
Take-Up
Rate
Difference
From Target
Take-Up
Rate
Uninsured FPL < 70 <.01 32.9 <.01 14.1
Uninsured 70 ≤ FPL ≤ 138 <.01 34.1 <.01 17.6
Uninsured 138 < FPL ≤ 250 <.01 25.0 <.01 12.5
Uninsured 250 < FPL < 400 <.01 14.7 <.01 6.1
Uninsured FPL ≥ 400 <.01 6.0 <.01 3.0

14.
13
CBO
Results

15.
14
CBO
Federal Subsidies for Health Insurance Coverage for People Under Age 65: 2019 to 2029 (May 2019), Table 1-1, www.cbo.gov/publication/55085
Health Insurance Coverage for People Under Age 65, 2019 to 2029

16.
15
CBO
Enrollment in Medicaid, 2016
Percent
Share of Medicaid Enrollees
(Full-year-equivalent full-benefit enrollment of the non-elderly)
As Estimated
by HISIM2
As Estimated
by the Treasury
(Form 1095-B)
Modified Adjusted Gross
Income as a Percentage of
the Federal Poverty Level
Under 100 55 53
100–150 20 20
150–200 12 11
200–250 7 6
250–300 4 3
300–400 2 3
400–600 1 2
Above 600 1 1
Total 100 100

17.
16
CBO
Enrollment in Employment-Based Coverage, 2016
Percent Employment-Based Coverage
Modified Adjusted Gross
Income as a Percentage of
the Federal Poverty Level
As Estimated
by HISIM2
As Estimated
by the Treasury
(Covered life years)
Under 100 7 6
100–150 5 5
150–200 7 7
200–250 7 8
250–300 9 8
300–400 15 16
400–600 21 22
Above 600 28 28
Total 100 100

18.
17
CBO
Marketplace Premium Estimates Compared With Actuals
Thousands of Dollars
Average Second-Lowest-Cost Silver Plan Gross Premium for a
Single 21-Year-Old for Health Insurance Purchased Through the Marketplaces

19.
18
CBO
“HISIM2—The Health Insurance Simulation Model Used in Preparing CBO’s
Spring 2019 Baseline Budget Projections” (presentation, April 2019),
www.cbo.gov/publication/55097
“HISIM2 Code Segments, April 2019” (supplemental data for “HISIM2—The
Health Insurance Simulation Model Used in Preparing CBO’s Spring 2019
Baseline Budget Projections”), www.cbo.gov/publication/55097#data
Sources and Preparation of Data Used in HISIM2—CBO’s Health Insurance
Simulation Model, Working Paper 2019-04 (April 2019),
www.cbo.gov/publication/55087
Health Insurance Coverage for People Under Age 65: Definitions and Estimates
for 2015 to 2018 (April 2019), www.cbo.gov/publication/55094
Federal Subsidies for Health Insurance Coverage for People Under Age 65: 2019
to 2029 (May 2019), www.cbo.gov/publication/55085
CBO Sources

20.
19
CBO
Kenneth E. Train, Discrete Choice Methods With Simulation (Cambridge
University Press, 2009), https://doi.org/10.1017/CBO9780511805271
Chieh-Hua Wen and Frank S. Koppelman, “The Generalized Nested Logit
Model,” Transportation Research Part B: Methodological, vol. 35, no. 7
(August 2001), pp. 627–641, https://doi.org/10.1016/S0191-2615(00)00045-X
Ithai Z. Lurie and James Pearce, “Health Insurance Coverage from Administrative
Tax Data,” Working Paper 117 (Office of Tax Analysis, February 2019),
https://tinyurl.com/y4fekr8h (PDF, 19 MB)
Outside Sources